EY re-envisions RAG around multimodal knowledge graphs to improve accuracy
What changed
EY has revamped retrieval-augmented generation (RAG) systems by integrating multimodal knowledge graphs. The update targets a key flaw in most RAG implementations that focus almost exclusively on pulling in text data. EY’s approach adds the ability to mine and link data from charts, tables, and other non-textual elements embedded in enterprise documents. This creates a richer, more complete knowledge base for large language models to draw from.
Why builders should care
Most enterprise data isn’t pure text. Critical facts often live inside spreadsheets, graphs, and structured data, which traditional RAG setups miss or mishandle. EY’s multimodal knowledge graph approach forces the AI to consider these data types as interconnected nodes, giving the model more precise and context-aware sources. For engineers and AI operators, this could mean more accurate decision support and fewer hallucinations when enterprise models generate answers.
The practical takeaway
Integrating multimodal knowledge graphs lets builders ground large language models in a fuller picture of enterprise data. This can lower the risk of model errors and improve trust in AI-driven outputs. For operators relying on RAG systems for analytics, risk assessment, or compliance, rethinking data ingestion to include charts and tables reduces gaps in model understanding. It also sets a precedent to move beyond text-only strategies toward more data-diverse architectures.
What to watch next
The big question is how broadly this approach will spread beyond EY. Watch for startups and enterprise AI product teams adopting multimodal retrieval frameworks. Also track advancements in tools that convert structured data into graph formats suitable for RAG. Expect a push for standards around indexing and querying multimodal knowledge to support more reliable AI grounding in complex corporate datasets.
AI Quick Briefs Editorial Desk
